first of all I want to say that I really enjoy working with Pytorch Geometric, it is very powerful and I appreciate the community growing around it. The recently introduced SparseTensor for modeling of adjacency matrices is really nice and helps me to work with far bigger graphs. However, it appears that the dropout_adj is not yet compatible with SparseTensor, thus my question if there is a smart trick to adapt for that (without converting the SparseTensor back to "normal" edge_index format) or at least a good pointer!
Hi,
first of all I want to say that I really enjoy working with Pytorch Geometric, it is very powerful and I appreciate the community growing around it. The recently introduced SparseTensor for modeling of adjacency matrices is really nice and helps me to work with far bigger graphs. However, it appears that the dropout_adj is not yet compatible with SparseTensor, thus my question if there is a smart trick to adapt for that (without converting the SparseTensor back to "normal" edge_index format) or at least a good pointer!
Cheers and thanks in advance!